Bull, Fr. Pêche Piscic. (2000) 357/358 : 227-240 — 227 —

GILL NET SAMPLING STANDARDISATION IN LARGE RIVERS OF FRENCH GUIANA (SOUTH AMERICA).

F.L. T E J E R I N A - G A R R O and B. DE M É R O N A

Laboratoire d 'hydrobiologie, Institut de Recherche pour le Déve loppement - Cayenne, Route de Montabo, B.P. 165, 97323 Cayenne Cedex, France.

Reçu le 20 janvier 1999 Received 20 January, 1999 Accepté le 01 juillet 1999 Accepted 01 July, 1999

ABSTRACT

In this paper is analyzed the sampl ing eff iciency of repeated batter ies of mixed gillnets mesh size f rom 15 to 70 m m . The study was conduc ted in three French Guiana rivers : the Sinnamary, Comté and Ma lmanoury rivers. Eff iciency of sampl ing methods is one of the problems faced by f reshwater ichthyologists. If the choice of one approach must consider its eff iciency and cost/prof i t rates for a 'de te rmina te level of accuracy, the abiotic characteristics of the env i ronment samp led has also an impor tant role in this process. Large rivers, like those samp led in French Guiana, present specif ic abiotic character ist ics that prevent the large use of non-select ive sampl ing methods like electrof ishing or rotenone poisoning. Differently, o ther sampl ing gear such as gi l lnets is used frequently. This kind of gear presents known bias problems related to f ish size or species. However, bias can be d imin ished through the use of mixed gil lnets mesh size, wh ich eff iciency was previously de termined. Results of Principal Componen t Analys is (PCA) suggest that gil lnet mesh size 15 to 35 m m are the most efficient in sampl ing fish fauna in the rivers sampled . Regarding the number of repeated batter ies, the analysis of Catch Per Unit of Effort (CPUE), species r ichness, sample structure (rank correlat ion) and the examinat ion of their variance suggest that the use of three batter ies gives a good est imat ion for the fract ion of the fish communi ty samp led .

STANDARDISATION DE L 'ÉCHANTILLONNAGE AU FILET M A I L L A N T DANS LES G R A N D S C O U R S D'EAU DE LA G U Y A N E F R A N Ç A I S E

( A M É R I Q U E DU SUD) .

RÉSUMÉ

Cette étude por te sur l 'analyse du compromis entre l'effort d 'échant i l lonnage et la minoration des variabi l i tés expér imenta les en util isant de façon répétit ive des batter ies de filets mail lants de 15 à 70 m m . Lé tude a été déve loppée dans trois cours d 'eau de la Guyane f rançaise : le Sinnamary, la Comté et la Malmanoury. L'efficacité des méthodes d'échanti l lonnage est un des prob lèmes majeurs dans le quot id ien des ichtyologistes. Le choix d'une méthode doit considérer son eff icacité et la relation coût/bénéf ice pour arriver à un certa in n iveau de préc is ion des résultats. Toutefois, dans ce processus les caractéristiques du mil ieu à échant i l lonner ont un rôle important . Les cours d 'eau, c o m m e ceux trouvés en Guyane f rançaise, présentent cer ta ines part icular i tés qui empêchent l'utilisation des méthodes non sélect ives c o m m e la pêche électr ique ou l'util isation d'ichtyotoxique. En revanche, d 'autres types de matér iel de pêche, c o m m e les fi lets maillants sont cou ramment uti l isés. Ce matér iel présente des problèmes de biais par

Article available at http://www.kmae-journal.org or http://dx.doi.org/10.1051/kmae/2001046

Bull. Fr. Pêche Piscic. (2000) 357/358:227-240 — 228 —

rapport aux espèces et à la taille des poissons. Toutefois, ce biais peut être réduit en utilisant des filets de différentes mail les, dont l'efficacité a été démontrée préalablement. Le résultat de cette étude suggère que les filets de maille 15 à 35 mm de vide de maille sont les plus efficaces pour échanti l lonner les peuplements de poissons des rivières ici considérées. Les résultats de l'analyse sur le nombre de batteries répétées basés sur la variance de la capture par unité d'effort, de la richesse spécif ique et de la structure de l'échantillon (corrélation de rang) suggèrent que l'utilisation de trois batteries représente un compromis acceptable entre l'effort d'échanti l lonnage et la précision de l 'estimation des caractérist iques du peuplement de poissons.

INTRODUCTION

One problem faced by many fish ecologists is to get unbiased est imates of fish assemblage in a studied habitat. A number of methods were developed (BACKIEL and W E L C O M M E , 1980 ; BAYLEY, 1985), but each of them is l imited in its use by constraints related to habitat characteristics.

In opening the « Large River Symposium » held in Honey Harbour (Ontario) in 1986, D.P. DODGE defined large rivers as those large enough to intimidate research workers (HYNES, 1989). Although it is obviously difficult to set an arbitrary boundary between small and large rivers, authors generally agree that above 1 m deep and about 20 m wide, a river can be considered as large (KELLERHALS and C H U R C H , 1989 ; C H U R C H , 1992). In these habitats the more efficient catching methods for fish are generally inapplicable. Direct observation by snorkell ing cannot be done because of the low water transparency. Trawling is impeded by frequent presence of rocks or woody debris at the bottom. Electric f ishing has been used in certain occasions (NELVA et al., 1979 ; PERSAT and COPP, 1990) but the method is l imited. In forest rivers, like those encountered in French Guiana, water conductivity is as low as 30 uS and electric f ishing is only possible through heavy equipment. Depth of many aquatic environments and fish behaviour make also this technique inadequate for large rivers (LAMARQUE, 1990). The use of ichthyotoxics, which is considered a very efficient method for catching the whole communi ty of f ish, is limited to small and shallow areas and cannot be used repeatedly. PONTON and COPP (1997) using the rotenone in Guiana rivers appointed that in deeper places the retrieval efficiency of sunk fishes decreases, greater water velocities require greater quantit ies of toxicants which can increase the risk of mortality of f ishes downstream the place sampled. Seine nets have been used in few occasions for biomass est imates purposes, but the method is restricted to relatively clean biotopes and disturbs the habitat (BAYLEY, 1982 ; LAGLER era/ . , 1971).

Finally gill netting remains one of the few available methods for catching fish in most large rivers. Naturally, fish ecologists know how uncertain this method is, mainly because of selectivity problems (HAMLEY, 1975) al though, it revealed efficiency in detecting changes in fish community structure of f loodplain lakes when compared to high efficient methods like electrofishing (TEJERINA-GARRO et at., 1998). The probability of capture of an individual fish using gillnets depends not only on the adequacy of its size to net mesh size (HAMLEY and REGIER, 1973 ; HENDERSON and W O N G , 1991 ; HELSER and CONDREY, 1991), but also on its shape and behaviour (HAMLEY, 1975 ; JENSEN, 1986). Individual capture is also submitted to chance. This situation leads to the statement thai getting an unbiased sample of a diverse fish assemblage in large rivers seems unrealistic. However, realistic is the possibility to compare samples trying to reduce the bias. Mesh size selectivity can be reduced using large range of gil lnets of different mesh sizes and using a large number of nets of each mesh size can reduce chance parameters. Obviously, there is a limit in the number of nets that a research team can handle. This

Bull. Fr. Pêche Piscic. (2000) 357/358 :227-240 — 229 —

study proposes a method to de termine the best compromise between the number of mesh sizes and the number of nets of each size, wh ich must be used in order to get a stabil ized sample in large rivers.

MATERIAL A N D M E T H O D S

The principle of the method used in this study was to set a large number of gil lnets in the same area within a l imited per iod of t ime. Gil lnets were organized in batteries ; each battery was composed by nine nets of mesh sizes 15, 20, 25 , 30, 35, 40, 50, 60 and 70 mm between adjacent knots. The choice of this combinat ion of mesh sizes was determined by earl ier sampl ing . A 10 m m gillnet has a very low efficiency, rapidly fi l led by debris or algae. Due to the scarci ty of very large species in French Guiana freshwaters, the capture of gil lnets of 70 m m mesh size is only acc identa l , mak ing the use of nets of larger mesh size inadequate. Each net was 25 m long and 2 m high. In order to get general results sampl ings took place in different per iods of the year and in different rivers.

Five samples were taken :

- In the S innamary River upst ream f rom the Petit-Saut reservoir (Figure 1). Sampl ing took place two t imes, in November 1995 with 5 batter ies and in November 1996 with 7 batteries. Sampl ing was done in three different- sites distant about 25 km one f rom other. Correlation of normal ized species abundance between the 3 sites are highly signif icant (Table I). In the three areas the river width var ied between 26 and 63 m and r iverbanks were covered by dense and cont inuous pr imary vegetat ion.

- In the S innamary River, downst ream from Petit-Saut d a m , sampl ing took place in April 1997 in a sect ion of 800 m at about 30 km f rom the dam (Figure 1). Five batteries were used. River width in sampl ing area ranged f rom 59 to 112 m. The river f lows through the coastal plain and is bordered by pr imary forest.

- In the Ma lmanoury River, a smal l coastal watercourse crossing the coastal p lain, one sample was taken in Apri l 1997 (Figure 1). Six batter ies of gil lnets were set in a section of about 800 m located in the middle course. In the sampled area the channel width varied f rom 25 to 35 m. This river runs throughout forest, savanna and some disturbed areas also.

- in the Comté River one sample was taken in March 1998. The sample took place in the most upst ream half of the river (Figure 1). Six batter ies were used. The river width ranges between 23 and 45 m and the banks are covered with vegetat ion.

Table I Correlation values a m o n g the three stat ions sampled in the upstream Sinnamary in 1995 and 1996. A = Saut Takari Tante, B = Saut Dal le, C = Saut Deux Roros. * n = 36, p < 0.01 ; ** n = 44 , p < 0 . 0 1 .

Tableau I Valeurs d e la corré la t ion ent re les trois s tat ions échant i l lonnées dans le Sinnamary amont en 1995 et 1996. A = Saut Takari Tante, B = Saut Dalle, C = Saut Deux Roros. * n = 36 , p < 0.01 ; ** n = 44, p < 0 .01 .

A

1995 Stat ion

B C A

1996 Stat ion

B C

A Station B

C

1 0.695* 0.690*

1 0.726* 1

1 0.646** 0.703**

1 0.680** 1

Bull. Fr. Pêche Piscic. (2000) 357/358 : 227-240 — 230 —

54° 53° 52°

Figure 1 Localization of the Sinnamary River (1), Comté River (2), and Maimanoury River (3) in French Guiana. Dots represents sampling sites. In Sinnamary upstream three sites were sampled (A = Saut Takari Tante, B = Saut Dalle, C = Saut Deux Roros) and one area downstream the hydroelectric dam of Petit-Saut (PS).

Figure 1 Localisation du fleuve Sinnamary (1), de la rivière Comté (2) et du fleuve Maimanoury (3) en Guyane f rançaise . Les points indiquent les sites d'échantil lonnage. Sur le fleuve Sinnamary trois sites ont été échanti l lonnés en amont (A = Saut Takari Tante, B = Saut Dalle, C = Saut Deux Roros) et un en aval du barrage de Petit-Saut (PS).

i ' 1 r~r

Bull. Fr. Pêche Piscic. (2000) 357/358 :227-240 — 231 —

The nets were set paral lel to shore line, in places with low current. They were deployed at 5 pm and retr ieved at 7 a m . Tentative of f ishing dur ing day showed that additional capture, in te rms of number of spec imens as wel l as species, is negligible.

Captured f ishes were identi f ied in f ield us ing keys for adul ts e laborated by ROJAS-BELTRAN (1984), S A N T O S et al. (1984), P L A N Q U E T T E et al. (1996), and BOUJARD et al. (1997).

In order to know the contr ibut ion of each gil lnet mesh size to total var iance a Principal Componen t Analys is (PCA) was conducted on the matr ix species f requency by mesh size for each sample using statistical package ADE-4 (TH IOULOUSE et al., 1997). Each column of mesh size gil lnets was t ransformed (standardized) and the PCA analysis was conducted using the covar iance matr ix. This kind of analysis and the t ransformat ion allows to compare heterogeneous var iables in nature and scale (SIMIER, 1998).

In order to determinate the opt imal number of gil lnets batteries, we generated randomly, without reposi t ion, batter ies composed by five nets of mesh size 15, 20, 25, 30 and 35 mm. Fifty random draws were made for each number of batteries used in each river sampled. The parameters examined were :

- Species r ichness - w e adjusted the me.an values of r ichness against number of batteries to a logistic curve : y = ya(1-e n ) in order to est imate the theoret ical species richness (Sra) for every river samp led .

- Catch Per Unit of Effort - number of spec imen captured/gi l lnet/hour.

- Sample structure using rank correlat ion between sample f rom each battery and total sample.

For each of these parameters the mean and the coeff icient of variat ion def ined in % were plotted against the number of batteries.

RESULTS

The contr ibut ion of each mesh size to total inert ia in the PCA analysis shows similar decrease of inert ia f rom mesh size 15 to 25 m m for the S innamary upstream in 1997, Sinnamary downst ream and Comté River. From mesh size 35 up, inert ia presents low values (< 4) until net 70 m m in the three rivers ment ioned (Figure 2). In the S innamary upstream in 1995, inert ia decreases f rom net 15 m m to 20 m m and reaches high value at net 25 mm showing low values (< 2) for meshes larger than 35 m m . In the Ma imanoury River inertia increases f rom gi l lnets 15 to 30 m m and reaches low values (< 2) f rom mesh 40 mm on (Figure 2) .

In addi t ion, mesh size 40 to 70 br ings few addit ional species : 3 at S innamary downstream stat ion, 2 in the Maimanoury River, 1 in the Comté River and none in the 2 samples f rom the upst ream part of the S innamary River.

There were col lected 4,121 spec imens distr ibuted in 74 species in the five stat ions sampled. 38 spec ies were samp led at the C o m t é River and 15 spec ies at the Maimanoury River. At the S innamary ups t ream 95, S innamary ups t ream 96 and Sinnamary downst ream the total of species col lected were 43 , 44 and 32, respectively (Table II). Adjusted curves of sample species r ichness versus number of batter ies show a clear decl ine in the s lope of three rivers sampled (Comté, S innamary upstream 96 and Sinnamary downst ream). In those rivers the percentage of the theoret ical species r ichness

Bull. Fr. Pêche Piscic. (2000) 357/358:227-240 — 232 —

(Sra) reached by four batteries is high : 78.8 % of 43.9 ; 82.1 % of 43.7 and 79.3 % of 2 7 . 1 , respectively (Figure 3). For the two other samples (Malmanoury and Sinnamary upstream 95), the adjusted curves appears like straight l ines and the number of species given by four batteries is low : 56.7 % of 15.7 and 59.1 % of 48.7, respectively (Figure 3). The coefficient of variation displays homogenous values f rom 3 batteries on . At this level, variance varies from 6.6 % in the Sinnamary upstream 96 to 13.1 % in the Malmanoury River, where the number of species is low.

15 20 25 30 35 40 50 60 70

Gillnet mesh size

Figure 2 Contributions of each gillnet mesh size to total inertia. C = Comté River ; M = Malmanoury River ; SU95 = Sinnamary upstream 95 ; SU96 = Sinnamary upstream 96 ; SID = Sinnamary downstream.

Figure 2 Contribution de chaque maille à l'inertie totale. C = Comté ; M = Malmanoury ; SU95 = Sinnamary en amont 1995 ; SU96 = Sinnamary en amont 1996 ; SID = Sinnamary en aval.

Plot of the mean rank correlation between the total sample and the samples from increasing batteries of gillnets, chosen randomly, are asymptot ic curves but no clear breaking in slope is evidenced. The coefficient of variation decl ines continuously f rom 1 to 6 batteries but reaches low values from 3 batteries where the coefficient of variation varies between 2.6 % (Malmanoury River) and 3.7 % (Sinnamary upstream 95). However, stabil ized values are observed at two/three batteries (Malmanoury and Sinnamary upstream 96) and at three/four batteries (Sinnamary upstream 95), but downs afterwards (Figure 4).

Bull. Fr. Pêche Piscic. (2000) 357/358 : 227-240 — 233 —

Table II

Frequency of species sampled at the Sinnamary, C o m t é and Malmanoury River,

French Guiana.

Tableau II

Fréquence des e s p è c e s échant i l lonnées sur le S innamary , la Comté et la

Malmanoury, Guyane f rançaise.

ORDER Comté Malmanoury Sinnamary Sinnamary Sinnamary Total

Family upstream upstream downstream

Genus and species 1995 1996

BELONIFORMES Belonidae Pseudotylosonjs microps 1 - - - - 1 CHARACIFORMES

Anastomidae Leporinus friderici 5 1 83 114 5 208

Leporinus despaxi - - - 4 - 4 Leporinus fasciatus - - 1 5 - 6 Leporinus gossei - 6 1 4 - 11

Leporinus granti - - 4 5 - 9 Leporinus pellegrini - - 2 3 - 5 Characidae Acestrorhynchus falcatus 13 14 8 60 - 95

Acestrorhynchus microlepis 13 - 25 136 4 178 Characidae

Astyanax bimaculatus - 3 2 1 - 6 Astyanax polylepis 1 - 2 13 2 18

Bivibranchia bimaculata - - 1 20 - 21

Bryconops melanurus - 31 61 - 92 Bryconops caudomaculatus 82 - 99 288 5 474

Bryconops affinis - - 6 38 - 44

Charax gibbosus 8 - - - 3 11

Hemigrammus uniiineatus - - 1 5 - 6 Moenkhausia collettii - - - - 2 2

Moenkhausia comma - 1 2 12 1 16 Moenkhausia georgiae - - 23 20 - 43

Moenkhausia oligolepis - - 7 83 1 91

Moenkhausia surinamensis 3 - 3 25 - 31 Piabucus dentatus 3 - - - 3 6 Poptella orbicularis 8 - 29 94 3 134

Triportheus rotundatus - - 191 54 13 258 Characidiidae Characidium sp. 1 - - 1 - - 1 Curimatidae Chilodus zunevei 2 - 1 27 - 30 Curimata cyprinoides 21 1 - 33 117 172 Cyphocarax spilurus 1 - 50 35.1 1 403

Cyphocharax hetleri - - - 33 - 33 Cyphocharax sp. 1 - 9 76 5 90 Erythrinidae Hoplias aimara 1 - 10 28 1 40

Hoplias malabaricus - - 2 20 - 22

Hemiodidae Hemiodopsis quadrimaculatus - 2 4 11 - 17

Hemiodus unimaculatus 55 - 15 - - 70 Lebiasinïdae

Pyrrhulina filamentosa _ 1 - - 1 Serrasalmidae Myleus rhomboidalis 1 - 1 3 - 5 Myteus temetzi 11 48 130 2 191 Serrasaimus humeralis 5 - - - - 5

Bull. Fr. Pêche Piscic. (2000) 357/358:227-240 — 234 —

Table II (Continued)

ORDER

Family

Genus and species

Comté Maimanoury Sinnamary

upstream

1995

Sinnamary

upstream

1996

Sinnamary

downstream

Total

CLUPEIFORMES Clupeidae

Pellona flavipinnis 1

Engraulidae Anchovia surinamensis 9

Anchoviella sp.

Lycengrauiis batesii 10

ELOPIFORMES Megalopidae

Megatops atlanticus

GYMNOTIFORMES

Hypopomidae Parupygus savannensis

Sternopygidae

Eigenmannia virescens

Sternopygus macrurus 1

PERCIFORMES

Centropomidae

Centropomus parallelus 1

Cichlidae Aequidens guianensis

Crenicichla saxatilis

Geophagus surinamensis

Hems severus 2

Satanoperca sp. aff. leucosticta 2

Scianidae

Plagioscion squamosissimus 3

Stellifer microps 8

SILURIFORMES Ageneosidae

Ageneiosus brevifilis 11

Ageneiosus sp. 4

Auchenipteridae

Auchenipterus nuchatis 40

Parauchenipterus galeatus 18

Parauchenipterus nodosus 4

Tatia intermedia

Callichthyidae

Cory doras spiturus 1

Doradidae

Doras carinatus 1

Loricariidae

Ancistrus hoplogenis 1

Ctenoloricaria macuiata

Harttia surinamensis

Hypostomus plecostomus 2

Lithoxus planquettei

Loricaria cataphracta

Loricaria gr. cataphracta 4

Pimelodidae

Pimelodella cristata • -

Pimelodus ornatus

Pseudopiatystoma fasciatum 1

Rhamdia quelen

83

72

2

14

1

74

10

94

2

3

4

2

1

1

5

2

4

78

59

290

2

5

2

35

2

238

2

1

2

44

2

10

21

3

6

154

2

3

8

80

4

662

103

77

4

7

2

4

10

1

Number of species

Number of specimens 38

358 15

200 43

870 44

2214

32

479

74

4121

Bull. Fr. Pêche Piscic. (2000) 357/358 ; 227-240 — 235 —

50 i

0 "1 ! 1 1 1 1 !

1 2 3 4 5 6 7

Number of batteries

Figure 3

Coefficient of variat ion (CV) and adjusted curves of species r ichness. Marks represent average of species r ichness. River codes are the same as those of Figure 2.

Figure 3

Coefficient de variat ion (CV) et courbes de richesse spécif ique ajustées. Les symboles représentent la moyenne de la r ichesse spéci f ique. Les codes des cours d'eau sont les m ê m e s que ceux de la Figure 2.

Bull. Fr. Pêche Piscic. (2000) 357/358 : 227-240 — 236 —

O H 1 1 1 1 1

1 2 3 4 5 6

Number of batteries

Figure 4 Coefficient of variation of Spearman correlation (CV) between total sample structure and those of batteries formed by mixed gil lnets mesh size randomly chosen. River codes are the same as those of Figure 2.

Figure 4 Coefficient de variation de la corrélat ion de Spearman (CV) entre la structure de l 'échantillon total et celle des batteries reconsti tuées aléatoirement. Les codes des cours d'eau sont les mêmes que ceux de la Figure 2.

The C P U E displays large variation f rom sample to sample but the mean is surpr is ingly well est imated by a unique battery. However, the var iance of this est imate from one battery is very high. The var iance downs, as the numbers of batteries increases and at 3 batteries on values are close. The coefficient of variat ion at this number of batteries varies between 15.8 % in the Malmanoury and 21.7 % at the Comté River (Figure 5).

Bull. Fr. Pêche Piscic. (2000) 357/358 : 227-240 — 237 —

300

UJ 250

ZJ 200

0 150 H 1 100

50

0

60 n

C M SU95 SU96 SID

Number of batteries

Figure 5

Coefficient of var iat ion of C P U E (CV) va lues by battery of gil lnet of the f ive sampling si tes. River c o d e s are the s a m e as those of F igure 2 .

Figure 5

Coefficient d e var iat ion d u C P U E (CV) par batterie de fi lets mai l lants. Les c o d e s des cours d 'eau sont les m ê m e s q u e ceux de la Figure 2.

DISCUSSION

In this study the contribution of each gillnet mesh size to total variance (inertia) decreases until still regularly, but the decrease is not similar for all rivers. Despite differences in number of batteries used in each river (Sinnamary upstream 96, 7 batteries ; Sinnamary upstream 95 and Sinnamary downstream, 5 ; Comté and Maimanoury River, 6 ) gillnets mesh sizes 15, 20, 25, 30 and 35 mm showed significant contribution to total inertia, whereas gillnets of mesh sizes 40, 50, 60 and 70 mm had low contribution. This

Bull. Fr. Pêche Piscic. (2000) 357/358:227-240 — 238 —

suggests that nets of mesh sizes 15 to 35 m m are more adequate to sample fish communi t ies in the rivers studied than nets of mesh sizes 40 to 70 mm. However, the inert ia variation observed among nets in the rivers sampled can also be explained partly by the characterist ics of the habitat (e.g. running velocity, kind of substrate, type of riverbank) rather than net catch (BOUJARD et al., 1990). A l though, other factors seems to be important. Indeed, tropical fish faunas, like those found in French Guiana, are naturally formed by a high proport ion of small species (< 10 cm) ( W E L C O M M E , 1979). This may explain the expressive contribution of net mesh size 15 mm in most of stat ions. However, the differences observed among gillnets 15 to 35 m m are also related to the abundance of some species. That is the case of Acestrorhynchus microlepis, Bryconops caudomaculatus and Cyphocharax spilurus in the Sinnamary upstream 96 (mesh size 15 mm) ; Bryconops caudomaculatus and Hemiodopsis quadrimaculatus in the S innamary upst ream 95 (mesh size 15 mm). Parauchenipterus galeatus is the main responsible for the peak observed at gillnet 30 m m in the Malmanoury River.

By the way, the low capture of large mesh size gil lnets (40 to 70 mm) would be due to f ish behaviour, especial ly that of large predators like Hoplias aimara and Ageneiosus brevifilis. These f ishes are almost always tangled in gil lnets of smal l mesh , where they are attracted by movements of caught f ish.

Because we are interested in fish communi t ies we did not consider here the biomass. Accidental capture of very large specimens introduce a great variabil i ty mainly generated by chance and so bringing an information of difficult ecological interpretat ion.

In spite of the problems related to the capture of f ish by gil lnets, our results are in accord with the suggest ion of HELSER and C O N D R E Y (1991) of mixing different mesh size in order to reduce bias due to selectivity of gil lnets. In our study the combinat ions of gil lnets mesh 15 to 35 m m seems to be satisfactory for sampl ing the f ish communit ies of the large rivers sampled.

In this paper the number of repeated set of nets was est imated based on two type of statistics (mean and variance of est imations randomly chosen in a system of increasing number of batteries of gillnets) f rom three parameters : C P U E , species r ichness and sample structure (rank correlation).

The est imation of the fish density (CPUE) does not vary much with increasing number of batteries in all the rivers sampled, but chance seems to be an important factor that accounts for the high variance observed among the samples. However, this variance may be also due to other factors such as seasonal variabil i ty (e.g., S innamary upstream 1995 and 1996) or river size (e.g., the Malmanoury, a coastal river compared to the other rivers sampled). CPUE results display that at 3 batteries of gil lnets relative low values is reached by variance, which afterwards displays little variat ion. This al low us to assume that from 3 batteries of gillnets on we get a good est imator of the density of the fish communi t ies sampled.

Species r ichness and sample structure do not stabil ize even beyond 5 batteries, but their evolution in function of the number of batteries is a lmost always asymptot ic. The coefficient of variat ion of the mean rank correlat ion displays plateaus f rom 2 to 3 or from 3 to 4 batteries in different samples and in all of them after 3 batteries var iance do not vary very much. This suggests that f rom 3 batteries on we have a good est imat ion of the sampl ing structure. Species r ichness var iance decreases fast f rom 1 to 5 batteries. However, f rom 3 batteries on, variance values vary slightly in all sampl ing. The adjusted points in relation to theoretical species r ichness appointed to an average of 71.2 % species sampled at 3 batteries of gil lnets. This represents an expressive number of the species of the fish communi ty sampled, even if we consider that species r ichness is a factor that

Bull. Fr. Pêche Piscic. (2000) 357/358 : 227-240 — 239 —

changes accordingly to the size of the sample ( M A G U R R A N , 1988) and it may be not wel l evaluated for the effort of sampl ing emp loyed in this study.

Despite the parameters used present conf idence limits rather than punctual results, these seems sensit ive enough to determinate the best number of batteries. Thus , 3 batteries seem to be a good compromise be tween sampl ing effort and accuracy of results that give us a rel iable image f rom the fract ion of the fish communi ty sampled . This number of batteries takes account of the mater ia l possibi l i t ies of a research team and the habitat density in a de te rmined sect ion of the river where the use of gi l lnets is feasible. However, because the data set used for this study does not cover the whole range of conditions in French Gu iana rivers, one could use an addi t ional battery providing a sufficient f ishing crew.

If the use of m ixed m e s h size of gi l lnets and repeated batter ies are useful to sampling fish communi t ies of large rivers, the pract ices of s tandardize this k ind of gear is not usual. Thus, s ince different r ivers present different specif ic composi t ion of the f ish fauna, it seems impor tant to include the standardizat ion of gi l lnets in sampl ing protocols.

A C K N O W L E D G E M E N T S

Special thanks to the t e a m of the Laboratoi re d 'Hydrobio log ie of the Institut de Recherche pour le Déve loppement ( IRD) - Cayenne for their suppor t on f ield and laboratory. W e are gratefu l to Pier Ange lo Tognini Fi lho, bache lor s tudent at the Department of Computa t ion Sc iences - Univers idade Catô l ica de Goiâs , Brazi l , for the statistical sof tware deve loped to random chose of numbers used in this paper. We thank two anonymous reviewers for their commen ts and suggest ions. F.L.T.G. received a grant from IRD.

REFERENCES

B A C K I E L T , W E L C O M M E R.L., 1980. Guidel ines for sampl ing f ish in in land waters. FAO Technical Paper No. 33 , 167 p.

BAYLEY P.B., 1982. Centra l A m a z o n f ish populat ions : B iomass, product ion and some dynamic character is t ics. PhD Thes is , Dalhousie University, 330 p.

BAYLEY P.B., 1985. Sampl ing prob lems in f reshwater f isheries. Proc. 4<h. Brit. Freshw. Fish. Conf., 3 - 1 1 .

BOUJARD T , PASCAL M., M E U N I E R F.J., 1990. Micro-répart i t ion spat io- temporel le du peuplement ichtyologique d 'un haut bassin f luvial de Guyane, l 'Arataye. Rev. Ecol. (Terre Vie), 45, 357-373 .

BOUJARD T., PASCAL M., M E U N I E R J.F., LE BAIL R Y , 1997. Poissons de Guyane guide éco log ique de l 'Approuague et de la réserve des Nouragues. Institut Nat ional de la Recherche Agronomique . Paris, 219 p.

CHURCH M., 1992. Channe l morpho logy and typology. In : P. C A L O W and G.E. P E T T S (Eds. ) , T h e r ivers h a n d b o o k : hydro log ica l a n d eco log ica l p r inc ip les . Volume 1 . Blackwel l Sc ience, Oxford , UK, 526 p.

HAMLEY J .M. , 1975 . Rev iews of gi l lnet select ivi ty. J. Fish. Res. Board Can., 32, 1943-1969.

HAMLEY J .M. , REGIER H.A., 1973. Direct est imates of gil lnet selectivi ty to Wal leye (Stizostedion vitreum vitreum). J. Fish. Res. Can., 30, 817-830.

Bull. Fr. Pêche Piscic. (2000) 357/358:227-240 — 240 —

HELSER T.E., CONDREY R.E., 1991. A new method of estimating gillnet selectivity, with an example for Spotted Seatrout, Cynocion nebulosus. Can. J. Fish. Aquat. Sci., 48, 487-492.

HENDERSON B.A., WONG J .L , 1991. A method for estimating gillnet selectivity of Walleye (Stizostedion vitreum vitreum) in multimesh multifilament gillnets in Lake Erie, and its application. Can. J. Fish. Aquat. Sci., 48, 2420-2428.

HYNES H.B.N., 1989. Keynote address. In : D.R DODGE (ed.), Proceedings of the International Large River Symposium (LARS). Can. Spec. Pub. Fish. Aquat. Sci., 106, 629.

JENSEN J.W., 1986. Gillnet selectivity and the efficiency of alternative combinations of mesh sizes for some freshwater fish. Journal of Fish Biology, 28, 637-646.

KELLERHALS R., CHURCH M., 1989. The morphology of Large Rivers : Characterization and management. In : D.R DODGE (ed.), Proceedings of the International Large River Symposium (LARS). Can. Spec. Pub. Fish. Aquat. Sci., 106, 31-48.

LAGLER K.F., KAPETSKY J.M., STEWART D.J., 1971. The fisheries of the Kafue River Flats, Zambia, in relation to the Kafue Gorge dam. FAO Fl : SF/ZAM 1 1 , Tech. Rep. 1, 161 p.

LAMARQUE P. 1990. Elecirophysiology of fish in electric fields. In : I.G. COWX and P. LAMARQUE (Eds.), Fishing with electricity : applications in freshwater fisheries management. Blackwell Scientific Publications, Oxford, 65-92 p.

MAGURRAN A.E., 1988. Ecological diversity and its measurement. CHAPMAN and HALL, London, UK, 179 p.

NELVA A., PERSAT H., CHESSEL D., 1979. Une nouvelle méthode d'étude des peuplements ichtyologiques dans les grands cours d'eau par échanti l lonnages ponctuels d'abondance. Comptes Rendus de l'Académie des Sciences, Paris, 289(D), 1295-1298.

PERSAT H., COPP G.H., 1990. Electric fishing and point abundance sampling for the ichthyology of large rivers. In : I.G. COWX (ed.), Developments in electric f ishing. Fishing New Books, N°27 Blackwell Scientific Publications, Oxford, UK.

PLANQUETTE P., KEITH P., LE BAIL P.Y., 1996. Atlas des poissons d'eau douce de Guyane (Tome 1). IEBG, MNHN, INRA, CSP, Ministère de l 'Environnement, Paris, 429 p.

PONTON D., COPP G.H., 1997. Early dry - season community structure and habitat use of young fish in tributaries of the River Sinnamary (French Guiana, South America) before and after hydrodam operation. Environmental Biology of Fishes, 50, 235-256.

ROJAS-BELTRAN R„ 1984. Clé simplifiée des espèces de Siluriformes. Bulletin de Liaison, Groupe de Recherche de Guyane, 63 p.

SANTOS G.M. dos, JEGU M., MÉRONA B. de., 1984. Catâlogo de peixes comerciais do baixo rioTocantins ; projeta Tucuruf. ELETRONORTE/CNPq/INPA, Manaus, 83 p.

SIMIER M., 1998. Initiation au logiciel ADE-4. ORSTOM, Centre ORSTOM de Montpellier. Montpellier, 83 p.

TEJERINA-GARRO F.L., FORTIN R., RODRIGUEZ M.A., 1998. Fish community structure in relation to environmental variation in floodplain lakes of the Araguaia River, Amazon Basin. Environmental Biology of Fishes, 51, 399-410.

THIOULOUSE J., CHESSEL D., DOLEDEC S., OLIVIER J.M., 1997. ADE-4 : a multivariate analysis and graphical display software. Statistics and Computing, 7, 75-83.

WELCOMME R.L., 1979. Fisheries ecology of f loodplain rivers. Lonqman, New York, 317 p.